How does ValiFit calculate town scores?

ValiFit scores every town in America on six composites — Safety, Education, Health, Environment, Appreciation, and Livability — using federally-published data, plus open Overture Maps place data for Livability. Each composite is a closed-form formula computed from federal inputs, then percentile-ranked against a state-tier peer cohort with a bootstrap 95% confidence interval. The six composite scores are blended into a personalized Match% using the weights you set during intake.

How does ValiFit calculate safety scores?

Safety is computed as safety_roi = (officers_per_1000 + 0.5) / (violent_crime_per_1000 + 0.5) per FBI UCR / LEE 2024 and FBI NIBRS 2024. The +0.5 additive (Laplace) smoothing handles zero counts defensibly. The town is percentile-ranked within its state-by-population-tier peer cohort, and the score is shown with a bootstrap 95% confidence interval.

How does ValiFit calculate education scores?

Education is computed as edu_roi_adj = (graduation_rate / (per_pupil_$ / 1000)) x (1 + (free_reduced_lunch_pct/100) x beta) per NCES F-33 and EdFacts ACGR. The free/reduced-lunch coefficient adjusts for student-body difficulty. The district is percentile-ranked within state-by-NCES-enrollment-tier peers with a bootstrap 95% CI.

How does ValiFit calculate health scores?

ValiFit computes health at two grains, and which one you are reading is stated on the page. At address grain, on a property page: health is a weighted average of outcomes (60%) and care access (40%), resolved at the property itself rather than at its county. Outcomes combine life expectancy (CDC/NCHS USALEEP, census tract) with the share of adults reporting fair or poor health (CDC PLACES, census tract), Gaussian-weighted from the property coordinate with a 2-mile decay and a 5-mile cutoff, so nearby tracts count more than distant ones. Care access is the geometric mean of three address-resolved measures: two-step floating catchment provider access, hospital capability (trauma level and beds divided by the population in each hospital catchment), and the count of distinct specialties reachable, all from CMS NPI 2024. The blended index is then percentile-ranked nationally. Outcomes are weighted above access because proximity to a hospital does not make a place healthy: a home can sit minutes from a trauma centre and still have a life expectancy over a decade lower than a suburb across the county line, and only the outcome measures separate the two.

How does ValiFit calculate environment scores?

Environment measures hazard exposure at the property, where lower exposure scores higher. Four hazard families are combined, each using a published federal severity standard rather than a weighting of our own: flood, from the FEMA flood zone the parcel actually falls inside (VE, AE, X and the rest, the same zones the National Flood Insurance Program prices against); natural disaster, from the FEMA National Risk Index, whose 18-hazard expected-annual-loss score covers wildfire, earthquake, hurricane, tornado and heat; radon, from the EPA Radon Zone; and toxic exposure, from proximity to EPA Superfund and brownfield sites, weighted so that distance matters. Flood and toxic exposure are resolved at the parcel itself, because two homes on the same street can sit in different flood zones. The National Risk Index and radon zone are published by county, and the response labels each family with the geography it came from, so a county-level figure is never presented as if it described your street. The four families are combined so that one severe hazard dominates rather than being averaged away by calmer ones, and the result is percentile-ranked nationally. The family weights and that blend are current defaults under review, not a federal standard, and they are the one part of this score we set ourselves.

How does ValiFit calculate appreciation scores?

Appreciation is the five-year FHFA House Price Index repeat-sales CAGR, EWMA-smoothed (decay 0.5) and percentile-ranked within the OMB-defined Metropolitan Statistical Area. FHFA HPI is the federally-published repeat-sales index that controls for property-quality drift in a way median sale price cannot.

How does Match% work?

Match% is the weighted geometric mean of the six composite scores: match_pct = product(composite_score_i ^ user_weight_i) ^ (1 / sum(user_weight_i)). Geometric mean (the same compensatory utility theory the UN HDI uses) prevents one high composite from masking a low one in a heavily-weighted dimension. The six composite percentiles are identical for every user; only the personal weight vector changes the blend.

Why is affordability not a composite?

Affordability is binary at the buyer level — either you can carry the payment or you cannot — so a continuous percentile score would be misleading. ValiFit enforces affordability as a hard-filter via /api/affordability, computing qualifying income from list price, current Freddie PMMS 30-year rate, county tax, insurance, and PMI. Locked per task #245.

What cohort does ValiFit use for percentile ranking?

Cohorts are federally defined: state by Census population tier (13 CFR Part 121) for Safety and Environment; state by NCES enrollment tier for Education; state by USDA RUCC county-tier for Health; OMB-defined Metropolitan Statistical Area for Appreciation. When a cohort has fewer than 30 peers, ValiFit expands outward (next tier, then state, then Census region) until N is at least 30. If even the widest cohort has fewer than 30 peers, the score returns null with reason insufficient_cohort — never a fabricated number.

How are confidence intervals computed?

Every score is published with a 95% confidence interval from a bootstrap of 1,000 resamples (deterministic seed for reproducibility). The bootstrap is mathematically equivalent to a Bayesian posterior under a uniform prior. For cohorts between 30 and 100 peers, ValiFit also applies James-Stein-style Bayesian shrinkage toward the state-median prior, tightening estimates for small samples while converging to the raw percentile as N grows.

Yes. ValiFit measures infrastructure investment (officers, providers, per-pupil spending, environmental hazards) — never demographic outcomes. No race, ethnicity, religion, family-status, or disability data enters any composite formula. The disparate-impact audit, applying HUD 24 CFR Section 100.500 analogously, is published in full at /trust/bias-audit: its most recent run (2026-05-16) tested 45 pairs — five scores (Education, Safety, Health, the price-appreciation score, and the blended overall score) against nine Census protected-class proxies — drawn from 24,306 municipalities, of which 17,944 carried both a score and the proxy and so entered each correlation. Eight pairs exceeded the |r| > 0.3 threshold, every one on the rank correlation (Spearman 0.305 to 0.515; Pearson at or below 0.178 on those same pairs). Environment and Livability are not covered by this audit.

Does ValiFit ever say a town is best?

No. ValiFit reports the data and computes a personalized match per source. The language used everywhere on the site is "matches you at X%" not "is the best." Every score carries its federal source citation inline, every cohort comparison is published, every formula is open. The user decides what matters.

Is real estate agent lead routing neutral?

No. Lead routing is neutral per ValiFit's constitutional charter at /trust/charters. No agent can pay for a ranking boost, no listing receives editorial promotion, and the scoring code is identical for every property regardless of whether the listing agent has a paid ValiFit account.

How often is ValiFit data refreshed?

Each source carries its own target refresh frequency, tracked per source in our data-lineage register: listings refresh continuously; the Freddie Mac PMMS mortgage rate weekly; FHFA HPI, BLS QCEW and BEA personal income quarterly; NCES F-33, EdFacts ACGR, FBI UCR and FHA loan limits annually; CRDC biennially. Several sources — including Overture Maps, CMS Hospital Quality and HRSA HPSA — are refreshed manually rather than on an automatic schedule. A target frequency is not a promise that a given source was refreshed on time, and we do not present it as one: the date that governs a score is the source vintage published next to it on the entity page.

Where can I see the scoring code and methodology spec?

The pre-registered methodology spec is published at /docs/SCORING_METHODOLOGY_v1.md and mirrored on GitHub at github.com/v4lifit-dotcom/valifit-scoring-methodology under CC-BY 4.0. The scoring implementation lives in lib/composites/unified-scorer.ts and lib/composites/math/percentile.ts. Any change requires a versioned release with changelog entry — the pre-registration is timestamped 2026-05-12 as prior art.

What happens when federal source data is missing for a town?

Per the omit-not-default rule, ValiFit never substitutes a hardcoded value while still attaching the federal source citation. A factor with missing source data is omitted from the composite, and the score is recomputed from the remaining factors. If the entire composite cannot be computed, it returns null with an explicit reason (insufficient_cohort, missing_primary_input, or biased_cohort) — and on the user-facing page that composite is marked unavailable, never faked.

Methodology · canonical scoring spec·6 composites · 2 hard filters · 1.7k LOC published

How ValiFit Scores Every Town in America

ValiFit publishes a town-by-town score for every incorporated place and active township in the United States. Each town is measured on six composites and filtered by two hard constraints. Every input is a federally-published value with a defined unit and a citable URL, with one stated exception: Livability is computed from Overture Maps open place data, which is not a federal source and is cited as such. Every score carries a bootstrap 95% confidence interval. The full pre-registered spec was timestamped 2026-05-12 and is published under CC-BY 4.0 — per FBI NIBRS 2024, NCES F-33, CDC NCHS, FEMA NFHL, and FHFA HPI as the primary federal sources.

FHA-safe (no demographic inputs)Pre-registered 2026-05-12CC-BY 4.0 methodologyBootstrap 95% CI on every score

1. The six scored composites

Each town receives a percentile score from 0 to 100 on five composites. Each composite uses one closed-form formula, one federally-defined peer cohort, and one bootstrap 95% confidence interval. No editorial weights live inside a composite — weights only appear at the user-controlled Match% layer.

Safety

Sworn-officer staffing relative to the violent-crime rate, Laplace-smoothed to handle zero counts.

Primary metric

safety_roi = (officers_per_1000 + 0.5) / (violent_crime_per_1000 + 0.5)

Formula. Higher safety_roi means more police presence per crime. The +0.5 additive (Laplace) smoothing is the defensible standard for count-based ratios where either numerator or denominator can be zero.

Cohort. State by Census population tier (13 CFR Part 121: micropolitan rural under 2,500 / small town 2,500-9,999 / small city 10,000-49,999 / mid city 50,000-249,999 / large city 250,000+). Expands to state then Census region when peer count is under 30.

Score rule. safety_score = ROUND(percentile_rank(safety_roi, cohort) x 100). Bootstrap 95% CI from 1000 resamples shown alongside every score.

Federal sources

Education

Graduation outcomes per dollar spent, adjusted for the difficulty of the student body (free/reduced-lunch share).

Primary metric

edu_roi_adj = (graduation_rate / (per_pupil_$ / 1000)) x (1 + (free_reduced_lunch_pct/100) x beta)

Formula. Outcome per dollar, adjusted for student-body difficulty. The beta coefficient is empirically derived from regression of per-pupil-spending residuals on F/R% across every NCES district (published in methodology-repo/verify/education-fr-coefficient.json).

Cohort. State by NCES district enrollment tier (under 300 / 300-2,500 / 2,500-10,000 / 10,000+ from NCES Common Core of Data).

Score rule. edu_score = ROUND(percentile_rank(edu_roi_adj, district_cohort) x 100). Bootstrap 95% CI shown alongside.

Federal sources

Health

Town grain, used on town profile pages. Geometric mean of provider access and life-expectancy outcome — both must be present for a strong score.

Primary metric

health_roi = sqrt( (providers_per_1000 / 2.65) x (life_expectancy / 78.8) )

Formula. 2.65 is the national average providers per 1,000 (CMS NPI 2024); 78.8 is national average life expectancy (CDC NCHS 2024). Geometric mean is the same family the UN Human Development Index uses — one near-zero input pulls the whole score down, preventing high access from masking poor outcomes (or vice versa).

Cohort. State by county-tier (rural / mid-rural / mid-urban / urban) from USDA ERS RUCC codes, matching the County Health Rankings methodology.

Score rule. health_score = ROUND(percentile_rank(health_roi, cohort) x 100). Bootstrap 95% CI shown alongside.

Federal sources

Environment

Town grain. Inverted percentile of federally-published contamination-source density near the town centroid. Flood is not an input to this score — a property page resolves flood zone at address grain; see the FAQ below.

Primary metric

env_score = ROUND(100 x (1 - percentile_rank(risk_count, state_tier_cohort)))

Formula. risk_count = count of EPA brownfield sites (44,437 nationally) + EPA Superfund NPL sites (1,837) within 5 miles of the town centroid, plus EPA AQS air-quality facilities (16,932) within 3 miles. Brownfields are roughly four fifths of it. The radius runs from the town centroid rather than the town boundary, so a large city can have in-boundary sources fall outside the radius and a small town can have out-of-boundary sources fall inside it. EPA SDWIS drinking-water violations and the two FEMA flood layers are NOT inputs to this town-grain count. Higher exposure count means lower score (inversion).

Cohort. State by Census population tier (13 CFR Part 121, same tier definitions used by Safety).

Score rule. Inverted percentile rank within state-tier cohort. Bootstrap 95% CI shown alongside.

Federal sources

Appreciation

Five-year FHFA House Price Index repeat-sales CAGR, EWMA-smoothed, ranked within the OMB-defined metro.

Primary metric

appreciation_cagr = (HPI_now / HPI_5yr_ago) ^ (1/5) - 1

Formula. FHFA HPI is the federally-published repeat-sales index — controls for property-quality drift in a way median sale price cannot (median price is contaminated by mix shifts). EWMA smoothing decay lambda = 0.5 weights recent years more heavily.

Cohort. Metropolitan Statistical Area (federally defined by OMB Bulletin 23-01). Expands to state then Census region when peer count is under 30.

Score rule. appreciation_score = ROUND(percentile_rank(appreciation_cagr, MSA_cohort) x 100). Bootstrap 95% CI shown alongside.

Federal sources

Livability

Category-weighted count of everyday amenities within three miles of the address, from open Overture Maps place data.

Primary metric

gravity_score = restaurants x 3 + coffee x 3 + bars x 3 + grocery x 3 + gyms x 2 + parks x 2

Formula. Places are counted within a 3-mile radius of the address and weighted by category, so a grocery store counts for more than a gym. The weights are fixed and identical on both the address-grain and town-grain read paths, which keeps the two commensurable. Livability is the only composite whose DIRECTION depends on your intake: an urban-seeking profile scores high amenity density as good, a rural-seeking profile scores it as bad, and a suburban profile treats it as mildly positive. The number is the same; what it means to you is not.

Cohort. Same peer cohort as the other composites — state by population tier, expanding to state then Census region when peer count is under 30.

Score rule. livability_score = ROUND(percentile_rank(gravity_score, cohort) x 100), then direction-flipped per the urban/rural preference set during intake. Data grain is disclosed per address: address-grain at 50 or more places within the radius, town-grain at 10-49, state-grain at 1-9.

Federal sources

Overture Maps Foundation — Places theme (open data)

2. Hard filters (Commute, Affordability)

Two constraints are intentionally NOT scored as composites — they are intrinsically user-specific and applied as hard-filters at match time. Locked per task #129 (Commute, 2026-05-29 with 2026-06-10 re-lock) and task #245 (Affordability).

Commute (hard-filter, not scored)

What. Commute is not a town composite. It is a per-user hard-filter applied at match time using your intake destinations and max-minutes thresholds.

How. commute_burden = sum (haversine_minutes_i x frequency_weight_i) / sum (frequency_weight_i). haversine_minutes = (haversine_miles x 60) / mode_speed where mode_speed in mph is car=35, transit=15, bike=12, walk=3 (BTS National Household Travel Survey averages). Towns exceeding your max-minutes for any frequent destination are filtered out before scoring.

Why hard-filter not composite. Commute is intrinsically user-specific — the same town can be a 12-minute drive for one buyer and a 90-minute drive for another. Putting commute in the geometric-mean blend would punish or reward towns based on a destination that has nothing to do with the town itself. Per locked task #129 (2026-05-29) and re-locked 2026-06-10 PM, commute stays a hard-filter, never a composite.

Federal sources

Affordability (hard-filter, not scored)

What. Affordability is also not a composite. It is a hard-filter on debt-to-income: towns whose median list price would push your DTI past the cap from your intake are filtered out via /api/affordability before scoring.

How. qualifying_income = monthly_PITI / (front_end_DTI_cap). PITI = principal + interest at the current Freddie PMMS 30-year fixed rate + property tax (county effective rate) + homeowners insurance + PMI when down payment is under 20%. FHA path applies if list price is at or below the county FHA loan limit; conventional path otherwise.

Why hard-filter not composite. Affordability is binary at the buyer level (either you can carry the payment or you cannot), so a continuous percentile score would be misleading. Per locked task #245, affordability is enforced as a hard-filter and not blended into Match%. The output is yes/no plus the qualifying income, monthly payment, and county FHA limit reference.

Federal sources

3. How Match% works

The six composite scores are blended into a personalized Match% using the weights you set during intake. The blend formula is the weighted geometric mean:

match_pct = product(composite_score_i ^ user_weight_i) ^ (1 / sum(user_weight_i))

Why a geometric mean rather than an arithmetic average: the geometric mean correctly penalizes imbalance under compensatory utility theory — a low value in a heavily-weighted dimension cannot be made up by a high value elsewhere. The UN Human Development Index uses the same approach for the same reason.

The six composite percentiles for any town are identical for every user — they are cohort-relative, not user-relative. Only the personal weight vector changes the blend. ValiFit cannot cherry-pick: the underlying truth about each town is invariant; the personal lens is yours to set. The match-percent code is in lib/composites/match-percent.ts and the property tests (monotonicity, fairness invariance, edge cases) live in __tests__/methodology-defensibility.test.ts.

Edge-case handling

Weight of zero on a composite excludes it silently (you opted out).
Null score on a composite excludes it with a warning; the remaining weight is redistributed.
Score of zero is floored at 0.5 so the geometric product is not annihilated.

4. Cohort and statistical method

Every score is a percentile rank within a federally-defined peer cohort. The cohort is never picked by ValiFit — it follows federal classifications so the same calculation can be reproduced by anyone with the source data.

Minimum cohort size

N ≥ 30 peers. If smaller, expand outward (next tier, then state, then Census region) until met.

Confidence interval

Bootstrap 1,000 resamples, deterministic seed for reproducibility, 2.5% / 97.5% empirical quantiles.

Bayesian shrinkage

For 30 ≤ N ≤ 100, James-Stein shrinkage toward the state-median prior. Converges to raw percentile as N grows.

Biased-cohort guard

If more than 50% of cohort peers are missing the primary metric, the remaining peers are self-selected — score returns null with reason biased_cohort.

Outlier handling

Tukey fences (1.5 × IQR) for cohort statistics. Outliers contribute to rank but do not distort the CI.

Time smoothing

EWMA with decay λ = 0.5 for time-series inputs (crime, grad rate, appreciation HPI). Recent years weighted more heavily.

Reference implementation: lib/composites/math/percentile.ts (percentileRank, bootstrapCI, expandCohort, computePercentileWithExpansion). Pre-registered spec: docs/SCORING_METHODOLOGY_v1.md.

5. Data sources

Every input to every score comes from one of these federally- or authoritatively-published datasets. Complete catalog with refresh cadence is at /trust/sources and /sources.

NCES F-33 (district finance)

Education

EdFacts ACGR

Education

NCES Common Core of Data

Education

CRDC Civil Rights Data

Education

CMS NPI / Hospital Compare

Health

CDC NCHS Life Expectancy

Health

HRSA HPSA

Health

County Health Rankings 2024

Health

FEMA NFHL flood zones

Environment (address grain)

FEMA NFIP flood claims

Environment (address grain)

EPA Facility Registry Service

Environment

EPA Underground Storage Tanks

Environment (ingested, not currently scored)

FHFA House Price Index

Appreciation

OMB MSA definitions (Bulletin 23-01)

Freddie PMMS mortgage rates

Affordability

CFPB Qualified Mortgage rule

Affordability

Census ACS B08303 commute

Commute

BTS National Household Travel Survey

Commute

Census TIGER/Line boundaries

Geography

Census Places (ANSI)

Geography

Overture Maps Foundation

POIs / Livability

6. What ValiFit will never do

These commitments are enforced in code, not in marketing copy. Each one corresponds to an iron-jaw antibody, a constitutional charter, or a federal-law constraint.

Never call a town "best" or "safest"
Per the Q7 match-not-claim doctrine, the language used everywhere on the site is "matches you at X%" not "is the best." Every score is presented with its federal source citation inline (for example, "per FBI NIBRS 2024") and its bootstrap 95% CI. Enforced by the no-claim-style-language-on-scores iron-jaw antibody.
Never use demographic data in scoring
No race, ethnicity, religion, national origin, family status (children), or disability data ever enters a composite formula. The Fair Housing Act and Equal Credit Opportunity Act prohibit demographic steering, and ValiFit measures infrastructure investment (officers, providers, per-pupil spending, environmental hazards) instead. The disparate-impact audit applying HUD 24 CFR Section 100.500 analogously is published in full, per pair, at /trust/bias-audit.
Never give agents preferential ranking
No agent can pay for a ranking boost. No listing receives editorial promotion. The scoring code is identical for every property regardless of whether the listing agent has a paid ValiFit account. Lead-routing neutrality is one of the six constitutional charters at /trust/charters.
Never fabricate a score
When federal source data is missing for a factor, the omit-not-default rule applies: the factor is dropped from the composite, the score is recomputed from the remaining factors, and if the entire composite cannot be honestly computed it returns null with an explicit reason (insufficient_cohort, missing_primary_input, or biased_cohort). On the page that composite is marked unavailable, never faked with a hardcoded default.
Never change methodology without versioning
The pre-registered spec is timestamped 2026-05-12 and any change requires a versioned release, a changelog entry, an iron-jaw audit, and a PR review. Methodology is locked; only the data refreshes on its federal cadence. The pre-registration timestamp is the prior-art shield against post-hoc tuning claims.
Never auth-wall a citable page
Town pages, property pages, methodology, trust pages, and the federal-source glossary stay free and crawlable forever per the IMDb-gating-policy lock (2026-05-17). Action-layer surfaces (DMs, tour booking, PDF dossier export, agent CRM) are the paid product — the public-citation surface stays open.

7. Transparency and source code

ValiFit publishes the spec, the implementation, and the audit trail. Anyone with federal source data can independently recompute any score on the site.

Pre-registered methodology spec

The full v1.0.0 spec — formulas, cohorts, edge cases, change protocol. Published under CC-BY 4.0.

Methodology repo (GitHub)

Public mirror of the spec plus verification scripts (PCA loadings, F/R coefficient regression, sample expected outputs).

Constitutional charters

Six public commitments — Source Sovereignty, FHA-Safe Methodology, Citation Density, Algorithmic Transparency, Public Bias Audit, Lead-Routing Neutrality.

Bias audit

Disparate-impact analysis published with raw data and audit code — every pair tested, the threshold, and which pairs cleared it.

Full source catalog

Every federal, state, and open-data source with refresh frequency and citation. Mirror at /sources.

Metrics catalog (canonical spine)

Every metric ValiFit reasons over with FHA-safety flag, direction-of-better, geographic level, and source citation.

Town directory

Browse every scored town. Each entity page surfaces the per-composite percentile, the source year, and the 95% CI.

Property scoring

Look up any address. The property page composes the six town composites with property-level FHA-safe attributes (price, beds, baths, sqft, year built, flood zone).

Citation: ValiFit, LLC. (2026). ValiFit Scoring Methodology v1.0.0. CC-BY 4.0. Available at github.com/v4lifit-dotcom/valifit-scoring-methodology. Pre-registered 2026-05-12. Last modified 2026-06-18.

8. Frequently asked questions

These answers are mirrored in a Schema.org FAQPage block on the same URL so AI search and rich-results crawlers can ingest them structured. Each answer carries its own federal-source citation where applicable.

How does ValiFit calculate town scores?: ValiFit scores every town in America on six composites — Safety, Education, Health, Environment, Appreciation, and Livability — using federally-published data, plus open Overture Maps place data for Livability. Each composite is a closed-form formula computed from federal inputs, then percentile-ranked against a state-tier peer cohort with a bootstrap 95% confidence interval. The six composite scores are blended into a personalized Match% using the weights you set during intake.
How does ValiFit calculate safety scores?: Safety is computed as safety_roi = (officers_per_1000 + 0.5) / (violent_crime_per_1000 + 0.5) per FBI UCR / LEE 2024 and FBI NIBRS 2024. The +0.5 additive (Laplace) smoothing handles zero counts defensibly. The town is percentile-ranked within its state-by-population-tier peer cohort, and the score is shown with a bootstrap 95% confidence interval.
How does ValiFit calculate education scores?: Education is computed as edu_roi_adj = (graduation_rate / (per_pupil_$ / 1000)) x (1 + (free_reduced_lunch_pct/100) x beta) per NCES F-33 and EdFacts ACGR. The free/reduced-lunch coefficient adjusts for student-body difficulty. The district is percentile-ranked within state-by-NCES-enrollment-tier peers with a bootstrap 95% CI.
How does ValiFit calculate health scores?: ValiFit computes health at two grains, and which one you are reading is stated on the page. At address grain, on a property page: health is a weighted average of outcomes (60%) and care access (40%), resolved at the property itself rather than at its county. Outcomes combine life expectancy (CDC/NCHS USALEEP, census tract) with the share of adults reporting fair or poor health (CDC PLACES, census tract), Gaussian-weighted from the property coordinate with a 2-mile decay and a 5-mile cutoff, so nearby tracts count more than distant ones. Care access is the geometric mean of three address-resolved measures: two-step floating catchment provider access, hospital capability (trauma level and beds divided by the population in each hospital catchment), and the count of distinct specialties reachable, all from CMS NPI 2024. The blended index is then percentile-ranked nationally. Outcomes are weighted above access because proximity to a hospital does not make a place healthy: a home can sit minutes from a trauma centre and still have a life expectancy over a decade lower than a suburb across the county line, and only the outcome measures separate the two.
How does ValiFit calculate environment scores?: Environment measures hazard exposure at the property, where lower exposure scores higher. Four hazard families are combined, each using a published federal severity standard rather than a weighting of our own: flood, from the FEMA flood zone the parcel actually falls inside (VE, AE, X and the rest, the same zones the National Flood Insurance Program prices against); natural disaster, from the FEMA National Risk Index, whose 18-hazard expected-annual-loss score covers wildfire, earthquake, hurricane, tornado and heat; radon, from the EPA Radon Zone; and toxic exposure, from proximity to EPA Superfund and brownfield sites, weighted so that distance matters. Flood and toxic exposure are resolved at the parcel itself, because two homes on the same street can sit in different flood zones. The National Risk Index and radon zone are published by county, and the response labels each family with the geography it came from, so a county-level figure is never presented as if it described your street. The four families are combined so that one severe hazard dominates rather than being averaged away by calmer ones, and the result is percentile-ranked nationally. The family weights and that blend are current defaults under review, not a federal standard, and they are the one part of this score we set ourselves.
How does ValiFit calculate appreciation scores?: Appreciation is the five-year FHFA House Price Index repeat-sales CAGR, EWMA-smoothed (decay 0.5) and percentile-ranked within the OMB-defined Metropolitan Statistical Area. FHFA HPI is the federally-published repeat-sales index that controls for property-quality drift in a way median sale price cannot.
How does Match% work?: Match% is the weighted geometric mean of the six composite scores: match_pct = product(composite_score_i ^ user_weight_i) ^ (1 / sum(user_weight_i)). Geometric mean (the same compensatory utility theory the UN HDI uses) prevents one high composite from masking a low one in a heavily-weighted dimension. The six composite percentiles are identical for every user; only the personal weight vector changes the blend.
Why is commute not a composite?: Commute is intrinsically user-specific — the same town is a 12-minute drive for one buyer and a 90-minute drive for another. Putting it in the scored blend would punish or reward towns based on a destination that has nothing to do with the town. Commute is applied as a hard-filter at match time using your intake destinations and max-minutes thresholds. Locked per task #129 (2026-05-29, re-locked 2026-06-10).
Why is affordability not a composite?: Affordability is binary at the buyer level — either you can carry the payment or you cannot — so a continuous percentile score would be misleading. ValiFit enforces affordability as a hard-filter via /api/affordability, computing qualifying income from list price, current Freddie PMMS 30-year rate, county tax, insurance, and PMI. Locked per task #245.
What cohort does ValiFit use for percentile ranking?: Cohorts are federally defined: state by Census population tier (13 CFR Part 121) for Safety and Environment; state by NCES enrollment tier for Education; state by USDA RUCC county-tier for Health; OMB-defined Metropolitan Statistical Area for Appreciation. When a cohort has fewer than 30 peers, ValiFit expands outward (next tier, then state, then Census region) until N is at least 30. If even the widest cohort has fewer than 30 peers, the score returns null with reason insufficient_cohort — never a fabricated number.
How are confidence intervals computed?: Every score is published with a 95% confidence interval from a bootstrap of 1,000 resamples (deterministic seed for reproducibility). The bootstrap is mathematically equivalent to a Bayesian posterior under a uniform prior. For cohorts between 30 and 100 peers, ValiFit also applies James-Stein-style Bayesian shrinkage toward the state-median prior, tightening estimates for small samples while converging to the raw percentile as N grows.
Is ValiFit FHA-safe?: Yes. ValiFit measures infrastructure investment (officers, providers, per-pupil spending, environmental hazards) — never demographic outcomes. No race, ethnicity, religion, family-status, or disability data enters any composite formula. The disparate-impact audit, applying HUD 24 CFR Section 100.500 analogously, is published in full at /trust/bias-audit: its most recent run (2026-05-16) tested 45 pairs — five scores (Education, Safety, Health, the price-appreciation score, and the blended overall score) against nine Census protected-class proxies — drawn from 24,306 municipalities, of which 17,944 carried both a score and the proxy and so entered each correlation. Eight pairs exceeded the |r| > 0.3 threshold, every one on the rank correlation (Spearman 0.305 to 0.515; Pearson at or below 0.178 on those same pairs). Environment and Livability are not covered by this audit.
Does ValiFit ever say a town is best?: No. ValiFit reports the data and computes a personalized match per source. The language used everywhere on the site is "matches you at X%" not "is the best." Every score carries its federal source citation inline, every cohort comparison is published, every formula is open. The user decides what matters.
Is real estate agent lead routing neutral?: No. Lead routing is neutral per ValiFit's constitutional charter at /trust/charters. No agent can pay for a ranking boost, no listing receives editorial promotion, and the scoring code is identical for every property regardless of whether the listing agent has a paid ValiFit account.
How often is ValiFit data refreshed?: Each source carries its own target refresh frequency, tracked per source in our data-lineage register: listings refresh continuously; the Freddie Mac PMMS mortgage rate weekly; FHFA HPI, BLS QCEW and BEA personal income quarterly; NCES F-33, EdFacts ACGR, FBI UCR and FHA loan limits annually; CRDC biennially. Several sources — including Overture Maps, CMS Hospital Quality and HRSA HPSA — are refreshed manually rather than on an automatic schedule. A target frequency is not a promise that a given source was refreshed on time, and we do not present it as one: the date that governs a score is the source vintage published next to it on the entity page.
Where can I see the scoring code and methodology spec?: The pre-registered methodology spec is published at /docs/SCORING_METHODOLOGY_v1.md and mirrored on GitHub at github.com/v4lifit-dotcom/valifit-scoring-methodology under CC-BY 4.0. The scoring implementation lives in lib/composites/unified-scorer.ts and lib/composites/math/percentile.ts. Any change requires a versioned release with changelog entry — the pre-registration is timestamped 2026-05-12 as prior art.
What happens when federal source data is missing for a town?: Per the omit-not-default rule, ValiFit never substitutes a hardcoded value while still attaching the federal source citation. A factor with missing source data is omitted from the composite, and the score is recomputed from the remaining factors. If the entire composite cannot be computed, it returns null with an explicit reason (insufficient_cohort, missing_primary_input, or biased_cohort) — and on the user-facing page that composite is marked unavailable, never faked.

Start your intake to see your Match% Trust framework All data sources